AI Security Issues Are Social Engineering Issues
Web page · https://oblomovka.com/wp/2026/11/11/ai-security-issues-are-social-engineering-issues
First seen by Oriel on .
AI summary
The essay frames prompt injection as social engineering applied to LLMs and argues capability-based security can limit the resulting damage, though it cannot eliminate human vulnerabilities.
- Prompt injection exploits models by smuggling instructions into their inputs, using tactics resembling impersonation, emotional manipulation, and gradual expansion of trust.
- Capability-based security gives each program only the authority required for its task, addressing the “confused deputy” problem and reducing an exploit’s blast radius.
- Existing capability-oriented projects include CHERI, Fuchsia, Capsicum, Cap’n Proto, Cloudflare’s Workers, Spritely Institute’s object-capability implementation, and OCapN.
- Adoption remains difficult because moving to capability-based security may require profound reengineering across the stack, from CPUs through operating systems and applications.
- The author hopes AI-assisted software rebuilding will accelerate this transition, while emphasizing that technical controls cannot prevent attackers from exploiting human flaws.
#prompt injection #capability security #social engineering #llm security
Generated by gpt-5.6-sol. Check the original for details.
Metric trends
Last 30 days · Up to 120 recorded samples per source and metric, with reposts kept separate · First and latest retained
Points
Score reported by the source.
- Lobsters: 0, latest observation ; -1 over observed period since
Comments
Comment count reported by the source.
- Lobsters: 0, latest observation ; 0 over observed period since
Aliases
| Source | External ID | URL |
|---|---|---|
| Lobsters | rehdv3 | https://www.oblomovka.com/wp/2026/11/11/ai-security-issues-are-social-engineering-issues/ Read discussion |